适用于Android平台的SG-YOLO人脸检测算法研究  

Research on SG-YOLO face detection algorithm for Android platform

作　　者：邹振超 刘瑜[1] Zou Zhenchao;Liu Yu(Faculty of Mechanical Engineering&Automation,Zhejiang Sci-Tech University,Hangzhou,Zhejiang 310018,China)

机构地区：[1]浙江理工大学机械与自动控制学院,浙江杭州310018

出　　处：《计算机时代》2023年第1期55-59,共5页Computer Era

摘　　要：现阶段对于人脸检测的研究主要在PC端,而基于移动设备如Android平台的人脸检测还在不断的改进之中。为了在Android平台提高人脸检测的速度和精度,在YOLOv4算法的基础上提出一个轻量级人脸检测算法SG-YOLO,在PC端训练好之后部署到RK3288主板的Android平台。实验结果表明,该算法在WIDER FACE数据集上有较高的准确率和检测速度,在Android平台可以满足人脸检测的实时性要求。For the current stage,the research on face detection is mainly on the PC,while the face detection based on mobile platforms such as Android platform is still improving.In order to improve the speed and accuracy of face detection on Android platform,a lightweight face detection algorithm SG-YOLO is proposed based on YOLOv4 algorithm,which is deployed to the Android platform with RK3288 motherboard after training on PC.The experimental results show that the algorithm has high accuracy and detection speed on the WIDER FACE dataset,and can meet the real-time requirements of face detection on Android platform.

关 键 词：深度学习 人脸检测 YOLO ANDROID 模型移植 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]